Research On Time-frequency Domain Identification Theories And Methodology Of Mechanical System Dynamic Characteristic Parameters

The time-frequency domain identification method of system dynamic characteristic parameters as new progress in the field of modal analysis and application of the time-frequency analysis theory is gradually attracting the attention of the international academia. On the basis of summing up the development of the modal analysis and the existing problems of the time-frequency domain parameters identification technique, this dissertation deals with the new identification methods of mechanical system dynamic characteristic parameters in time-frequency domain and the key issues in the process of parameter identification. The following contributions have been made.1) In order to solve the problem of detecting a time-frequency image ridge, the particle swarm optimization (PSO) algorithm is introduced. A method of single ridge detection based on the standard PSO algorithm is proposed. Furthermore, a method of multi-ridge detection based on the improved multi-objective PSO algorithm is presented by modifying the PSO evolutionary equations. The simulation experiment results prove that these methods are fast and simple, with high accuracy, less iterations and small noise.2) A hill-climbing algorithm is proposed for detecting multi-ridge. Experiment results show that the method is effective and accurate for multi-ridge detection, and those virtual ridges are eliminated by setting up the ridge energy threshold and the distance among ridges.3) Starting from the model of impulse response of multi-degrees of freedom (MDOF) system, an identification method for the system modal parameters with the time-frequency kernel is formulated. By firstly detecting the ridge using the CWD (Choi-Williams Distribution)-Time-frequency Reassignment and the improved Hough transform and then evaluating the inherent frequency, the integrated identification process is established. The integrated identification process of the method is established. The simulation results show the identification method not only has a higher resolution and is not sensitive to noise, but also is characterized with short time of computation.4) To expand the scope of the application of the wavelet identification, an identification method combining the NExT (Natural Excitation Technique) and the continuous wavelet transform is proposed. The equations of the parameters estimation using the wavelet identification method are derived, and the whole process of the parameters estimation is given. The selection of basic wavelet is fully studied. The advantages, with which the Gabor wavelet can minimize the edge-effects and improve the capability of identifying the heavily coupled modes, are systematically analyzed. Thereafter, the influences of Gabor mother wavelets with different shaping factors on the accuracy of the modal parameter identification are evaluated. The robustness of this method is proved by simulation.5) An EMD (Empirical Mode Decomposition) cooperated with adaptive matched pursuit identification method is proposed. The Laplace wavelet is selected for time-frequency atom to formulate the four parameters time-frequency atom dictionary. The adaptive feature matching algorithm based on PSO algorithm is presented. The simulation results show that after the pretreatment of the EMD, the speed and accuracy of parameter matching are significantly upgraded.6) These algorithms are applied separately to socle beam testing, and the feasibility as well as accuracy of these methods are justified by technical practice.